Rail Road Car Wheel Contamination Detection

Fall 2015—Spring 2016

A hump yard is a type of classification yard used to

separate railroad cars onto one of several tracks

leading to different destinations. During the sorting

process, cars are decoupled and retarders — essen-

tially disk-brakes — are used to slow down cars for

when they recouple.

If contaminants are present on the wheels, the re-

tarders won’t operate effectively and the cars can

blow apart couplers or even derail.

In order to investigate possible detection methods,

the effects of different contaminants on the system

needs to be quantified.

A test stand was designed to simulate the hump yard re-

tarder breaking system in order to get frictional data and

gain insight into detection methods. The test stand has a

3 HP motor with a 19.18:1 gear reducer to get the wheel

spinning at approximately 10 mph. The variable frequen-

cy drive, which runs the motor, and pneumatic braking

system are controlled using LabVIEW.

Problem Contaminant Test Stand

Team Members (Left to Right)

Joel Yauk, Wayne Helminen, Joshua Manela, Dylan Etelamaki

Advisor: Dr. Duane Bucheger

Results

The data shows that even thin films of the tested contaminants result

in substantial reductions of the coefficient of friction. This is a major

concern at the hump yards because a contaminant that cannot be

seen could go unnoticed, while still reducing the coefficient of friction

enough to cause problems. Contaminant detection methods will have

to be precise enough to detect very thin layers of contaminants. Be-

cause of this, visual detection methods would be inadequate.

Investigation

Members of the group visited a Norfolk Southern hump

yard in order to speak with yard personnel and get a

more hands-on perspective of the problem.

Finding the Coefficient of Friction

Two load cells are used in the brake assembly, one inline

with the brake shaft to measure the normal force to the

wheel and one perpendicular to the brake shaft to measure

the frictional force. The coefficient of friction can then be

calculated using the two force values.

Preliminary Encoder Detection Method with an SVM

The braking system uses two solenoids, which when energized

apply 100 psi of braking force to the wheel. A rotary encoder is

also mounted on the wheel giving a speed reference.

It was noticed that there was more acceleration jerk on a clean wheel than on a contaminated wheel. A

Support Vector Machine (SVM), a commonly used Machine Learning classifier, was fed frequency data

from the encoder. In preliminary testing, leave-one-out cross validation found 70% accuracy in detection

(test accuracy). These results show possible merit in exploring jerk-based detection methods.

Fusing the above data with sound and/or thermal data could increase the accuracy of the classification.

This allows sensor-fusion Machine Learning methods to be investigated.